Journal of Economic Behavior & Organization Vol.

61 (2006) 668–690

Organizational performance in hierarchies and communities of practice
Olivier Dupou¨ et a , Murat Yıldızo˘ glu b,∗

BETA (UMR CNRS 752) Universit´ e Louis Pasteur, 61, Avenue de la Forˆ et Noire,67000 Strasbourg, France b E3i, IFReDE-GRES, Universit´ e Montesquieu Bordeaux IV, Avenue L´ eon Duguit, 33608 Pessac, France Received 26 October 2003; accepted 21 July 2004 Available online 24 July 2006

Abstract In an earlier article we studied the Communities of Practice and their conditions of emergence using an Agent based model with a set of agents facing a continuous flow of problems. We now center our analysis on the performance of this organizational structure compared to a two-level hierarchical delegation structure. Our results show the crucial role played by the communication and the specialisation of agents; especially that community structures are efficient for competence building and learning in the long term. This paper backs the claim made by (Bowles, S., Gintis, H., 2002. Social capital and community governance. The Economic Journal 112, 419–437.) that hierarchy and communities are complementary modes of governance. © 2006 Elsevier B.V. All rights reserved.
JEL classification: D2; D83; L2 Keywords: Communities of practice; Learning; Emergence of networks; Organisational efficiency; Hierarchy

1. Introduction The idea that informal social structures play an important role in the behavior and capabilities of organizations is a long-standing idea in disciplines such as the sociology of organizations (Crozier and Friedberg, 1977) or management studies (Brown and Duguid, 1991, 1998). The studies of such informal structures have attained a growing interest with the recent emphasis put on the knowledge-based economy and the collective learning processes that become the key factor for success of firms in such a framework. Indeed, an important part of the learning processes within organizations are increasingly seen as taking place within the informal networks nested in them.

Corresponding author. Tel.: +33 556 84 54 53; fax: +33 556 84 86 47. E-mail addresses: (O. Dupou¨ et), (M. Yıldızo˘ glu).

0167-2681/$ – see front matter © 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.jebo.2004.07.011

O. Dupou¨ et, M. Yıldızo˘ glu / J. of Economic Behavior & Org. 61 (2006) 668–690


In particular, the concept of community of practice is deemed particularly useful by a number of scholars to account for the learning processes taking place within an organization (Brown and Duguid, 1991, 1998; Wenger, 1998). The concept of community of practice was introduced by Lave and Wenger (1990) who, by focusing on individuals’ practices, identified groups of persons engaged in the same practice communicating regularly with one another about their activities. Members of a community of practice essentially seek to develop their competencies in the considered practice. Communities of practice can then be seen as a means to enhance individual competencies; they are oriented toward their members (Lave and Wenger, 1990; Brown and Duguid, 1991). This goal is reached through the construction, the exchange and the sharing of a common repertoire of resources (Wenger). According to this definition, communities of practice appear as important loci for competence building within firms. As Wenger puts it, communities of practice are elementary regimes of competence. Amin and Cohendet (2003, 74) note “These communities might be found in traditional work divisions and departments, but they also cut across functional divisions, spill over into after-work or project-based teams, and straddle networks of cross-corporate and professional ties. For example, within firms, classical communities include functional groups of employees who share a particular specialisation corresponding to the classical division of labor (e.g. marketing or accounting). They also include teams of employees with heterogeneous skills and qualifications, often coordinated by team leaders and put together to achieve a particular goal in a given period of time”. However, although some works, either empirical or theoretical, investigate the impact of communities of practice upon the performance of the organization as a whole (Orr, 1990; Huberman and Hogg, 1995), such studies remain rare. More precisely, some measures of performance of communities themselves would be desirable. Moreover, the interplay between communities and hierarchy and the output of this interplay deserves deeper analysis. These issues are difficult to tackle in traditional ways since the impact of communities upon hierarchical structures and their potential benefits in terms of performance remain largely invisible in case studies. Their existence can be evidenced, but the true input of their activities for the firm are hard to assess. Besides, most of the theoretical work on communities of practice leaves some questions (such as the boundary of communities, for instance) unanswered. The concept is thus difficult to formalize using, for example, a mathematical apparatus. In a previous work (Dupou¨ et et al., 2003), we explored the emergence of communities of practice from a collection of individuals engaged in problemsolving activities. We evidenced sufficient conditions for the emergence of such social structures and show the usefulness of some indicators to characterize networks of communities of practice. This paper goes one step further and, using computational methods, seeks to shed some light on the role of communities in the performance of organizations, using computational methods. Even if this article’s main contribution concerns the emergence and the efficiency of the communities of practice, we use the hierarchy as a benchmark case. As a consequence this article is somewhat connected to the literature that has been mainly developed following the seminal article of Bolton and Dewatripoint (1994). This is not the place to review this quite rich literature, but we can emphasize the complementarity that exists between our work and these articles on the efficiency of hierarchy. Bolton and Dewatripoint consider the efficiency of different structures for the communication networks in an organization that must handle a continuous flow of information. They study the trade-off that must be established between the specialisation of the agents and the cost of communications under different–given–network structures and characterize the conditions under which a pyramidal structure can become the most efficient one. Even if the main problem of this article is not very far from some of the points we study in our model, the structure of these models are very different. In our case the structure of the communication

the results in terms of comparative performances of organizational structures and the specific impact of learning and communities are considered. within the latter. The last part of the article uses regression trees to analyze the role of the parameters of the model using the Monte Carlo method. The first line of work is quite far from the focus of our article.u-bordeaux4. but it results from the interactions between agents. two types of structures are contrasted: a pure network of communities of practice and a hierarchy in charge of the division of tasks. We will reconsider some of the results of Bolton and Dewatri-point later in the article when we expose ours. Each problem is randomly chosen from the problem space and is sent to an agent chosen randomly. By resorting to computer simulations based on the multi-agent system paradigm. We show that under some conditions. which we will discuss later in the article). With a very different approach. The model The model we present here1 is divided in two parts. this contribution explores the performance of various organizational settings and.html. this is the first article that uses regression trees for analyzing the results of simulations. but the general approach and the main point of attack of our article is quite different from this literature where the structure of the network is generally given (it does not result from the learning and the interaction of the agents) and the results depend on the particular cost function used to evaluate the burden of the communications. Moreover these articles are not interested in the quality of the decisions taken using the treated information (except Visser. It is then possible to compare the outcomes of each of these organizational settings. The phone calls received by the members of a help desk could correspond to such a process. If the problem does not belong to the realm of specialisation that the agent has built over time through learning 1 The javadoc information about the classes used in our program can be consulted at the following web site: http://beagle. This work is organized as follows. Yıldızo˘ glu / J. Mainly. The second line is more connected to our work. this structure can take the form of a communities of practice. Other articles are also dedicated to the search of an efficient hierarchy structure for organizations. First the different components of the model (the role and capabilities of individual agents and the overall structures) are presented. They consider the hierarchy as a distribution of authority in the organization (like Hart and Moore. Further. The first one (that we call communities of practice and CP hereafter) is a collection of agents having to solve a flow of problems and endowed with the abilities to learn by themselves or by interacting with one another. 2000. M. The development of finer databases should allow a stronger convergence between the theoretical literature and econometric studies.670 O. 2000). Visser. Delmastro (2002) tries to evaluate the determinants of the hierarchical structure in Italian plants econometrically but neglects many dimensions studied by the theoretical literature. of Economic Behavior & Org. probably because of the difficulty of observing these dimensions empirically. in particular. the role of communities in the performance of a firm. 2005) or as a structure of the communication channels (like Bolton and Dewatripoint. Dupou¨ et. 2. 61 (2006) 668–690 network is not given. one can either authorize the potential emergence of communities by allowing the communication between agents or remain in a strictly top-down decision process. .fr/yildi/hiercp/index. but also Garicano. The qualitative results obtained using the Monte Carlo method are very robust. To our knowledge. Next. but our article attacks these questions using a totally different approach. Their introduction corresponds to a contribution of this work at a methodological level.

The article questions the various dimensions of these different organizational settings with a particular interest in the role of communities in the global performance of a firm. We will discuss these elements in more detail later in this section. 50 or 75% (respectively coded as actions 0–5. Such a situation corresponds to a loss for the firm and hence a decrease in performance. That could make the classification more difficult. Otherwise it accepts the problem and tries to give an answer using its rule set. Formally. In addition. the problem remains unsolved (the system does not have the competencies to deal with this specific problem). and each binary string represents a particular situation and occupies a position on that grid. 2000). Yıldızo˘ glu / J. agents can communicate with each other. An action is associated with each situation (for the agent. and one of the environments we have tested corresponds to rectangular surfaces. we consider two possibilities: either agents are able to communicate horizontally or not. M. The linear nature of the complementarity that exists between these two dimensions of the projects implies smaller surfaces when we get closer to the origin. Ouksel and Vyhmeister. for example). we suppose that a project is characterized by two criteria stocked as a eight bits long binary string. communities of practice are susceptible to emerge. for example Carley and Lin. 20. see Fig. 30. hereafter) since the manager has to ask its subordinates for advice in order to answer the problems submitted by the environment. For each enterprise.1. we present the problem space that organizations are facing. 2. 61 (2006) 668–690 671 processes. 1). The second part of the model consists of an hierarchical structure where a manager is added to the community of the agents. In this article. If the last agent in the list cannot provide any answer.O. The problem space is thus a grid. The linear complementarity relation better corresponds to the nature of the problem we study (riskiness vs return of the projects. The first four bits code the value of the first criterion and the four last bits the value of the second one. The role of the manager is to receive the problems from the environment and to select the agent it deems the most able to solve this problem. In their endeavor to determine the optimum level of warranty. In this way. Its role is thus to carry out a division of labor. The overall dynamics thus result both from individual learning by doing and learning by interacting between agents. 10. In the rest of this section. the various possibilities take the form [condition:action]. it passes the problem to the next agent in the random list. The structure we have retained for this article does not imply any noticeable difficulty for learning in comparison with the rectangular one. Such problems are quite common in organizational contexts (cf. According to each particular situation. the agent must propose an amount of warranty (the action): for example. 1997. 40. They can thus exchange the “best practices” concerning a particular subspace of the problem space. the learning processes of agents. If the agent does not receive any response from the contacted agents. it consults the community that it has constructed over the past interactions. it refuses the problem and passes it to the next agent in a randomly constructed list of agents. the simulation model is inspired by a specific empirical context: a population of agents whose task consists in allocating financial warranties for bank loans to small enterprises. The environment The agents in the model face a continuous flow of classification problems. where the condition is the binary string and the action is an integer). We called this part of the model hierarchy with delegation (HD. . If it does not possess a satisfactory rule. Dupou¨ et. the implemented communication processes and the role of the manager. of Economic Behavior & Org. a given agent has to determine a certain amount of warranty by considering a set of criteria.

The task of the agents is then to determine the amount of warranty associated with each situation. This process corresponds to the individual learning process. a* is the correct answer expected by the environment and a the answer proposed by the agent. Dupou¨ et. An XCS receives a signal (problem) from the environment through a detector (Fig. The reward is computed as follows: f (a) = ∗ e(|a −a|) C (1) where C is a constant. The agent uses the reward it receives from the environment to actualize and refine its behavior. 1. of Economic Behavior & Org. Our article is not the first one to use LCS to represent adaptive individual learning. an algorithm from the family of LCS2 . Moreover. a LCS is a relevant tool for simulating relatively realistic learning processes (see Lanzi and Riolo. Yıldızo˘ glu / J. 1995). 1989). This signal is then compared to the set of rules the agent has in its memory. a rather natural way to represent this kind of process is the learning classifier system (LCS). the prediction of performance. 2). The problem space. it proposes a value greater than the optimal one. the error associated to the prediction 2 A very detailed algorithmic description of this classifier system can be found in Butz and Wilson (2000. These rules are made of a condition part and an action part.672 O. and given the multi-agent system we adopt. several parameters are associated with each rule: p. Its experience (the results it obtained from past trials) must help it in refining its judgement. The gap between the correct amount of warranty and the answer proposed by the agent determines the reward of the agent. 61 (2006) 668–690 Fig. The reward hence takes into account these two types of costs.2001). If. 2. Modeling of individual learning Each agent has to classify various situations (projects represented as binary strings) in one of the six areas defined by these six possible actions (see Fig. the loss would be greater than for the correct answer.2. Schulenburg and Ross (2000) uses LCS to study the performance of an artificial stock market on real data. then in case of bankruptcy of the warranted firm. ε. M.. it does not obtain the full amount of commissions that it could have expected from the situation. If the agent proposes an amount of warranty lower than the optimum. the structure we adopt for each individual agent is an XCS (Wilson. This is a typical classification problem. on the contrary. . For example. Indeed. More specifically. 1). 2000 and Holland et al.

2. we endow the agents with the following characteristics. Yıldızo˘ glu / J. 61 (2006) 668–690 673 Fig. Based on the literature on communities of practice. the agent can select the action predicting the highest reward from the environment or use a roulette wheel process to select the action to be fired. and this probability increases with the performance of the rule. The environment then rewards the XCS that uses this reward to adjust the parameters of its rules. For each problem. In our setting these two values of selectAg (0. Moreover. the corresponding value in the prediction array is nil. Schematic XCS (adapted from Wilson). each rule has a positive probability of being chosen. the best action is chosen.1) respectively correspond to agents who favour exploration (0: roulette wheel) or exploitation (1: best action). In the latter case. An agent seeks to develop its competencies on a given practice that does not constitute its entire activity but is nonetheless deemed critical by the agent. This possibility is behind the capacity of the Classifier System to generalize from observations. an action is chosen and triggered in the environment. M. This array contains for each possible action the average relative performance (fitness × prediction) of the rules that prescribe this action in the match set. A prediction array is then constituted in order to evaluate the reward associated with each advocated action. In our implementation. they are introduced into the population. rules regarding which the condition part fits the signal from the environment are stored in an array called the match set. The Match set is constructed using the rules that contain the signal 0011 in their condition part. When the genetic algorithm discovers new rules. the roulette wheel is used. Dupou¨ et.O. A parameter selectAg controls the choice of the selection mode: when set to 1. If no rules in the match set prescribe a particular action. The sign # corresponds to a joker that can either match 0 or 1. when set to 0. the fitness that measures the accuracy of the rules (a function of the inverse of the error) and that is used in the genetic algorithm used by the XCS to explore the problem space. of performance. F. the learning . of Economic Behavior & Org. Based on this array.

3). 2. . 3 This distance is normalized by the maximal possible distance in the environment. The requesting agent collects all answers and evaluates them. the agent becomes highly specialised. the agent copies it in its memory and uses it to answer the signal from the environment. the agent is not specialised and can accept all the problems submitted by the environment. • If predictions are equivalent (either several rules it received have identical prediction. When these parameters are not discriminating. communication. compRange also intervenes in the mutations in order to impede their capacity to disturb the learning process of the agent (the Genetic Algorithm is niche-based in the specialisation area of the agent). but when it is close to 0. the agent computes the maximal relative distance3 between the condition part of the best rules in its classifier set (the ones with a fitness better than the average of its classifier set) and the submitted problem. The variable compRange corresponds to the maximal possible distance between the specialisation area of the agent and the submitted problem. common to all agents). • In case of equivalence between prediction errors. then it adopts it. In the following periods.3. For each submitted problem. the agent chooses a rule at random. it engages a communication process by requesting help from other agents in its community. The surface of the specialisation is controlled by the parameter compRange that can rank from 0 to 1. determines whether or not the agents have the ability to communicate. This address book is originally empty. It then stores in its address book the references of agents answering him (not only the ones giving him the best answers).1] as compRange. or the rules it received and the ones it already possesses have identical prediction).674 O. • If a received rule offers a better prediction of performance. If an agent does not possess a satisfying rule to answer a problem but the problem lies nonetheless in its realm of specialisation. We now present the communication process (see Fig. the agent will question the agents present in its address book. This parameter can take the values 0 or 1. The agent can have no rule satisfying the signal from the environment or have only a poor rule (with a prediction of performance below an acceptable threshold. Learning by interacting In our implementation. it compares the prediction error and opts for the rule having the lowest one. By accepting a rule. Each contacted agent returns the best matching rule it has in its classifier set. M. As a consequence. 61 (2006) 668–690 process of each agent is oriented: new rules produced by the genetic algorithm are inserted in its memory only if they are in its realm of specialisation. it asks a percentage (consultRatio) of the whole population at random. it compares values of fitness and chooses the rule with the highest fitness. Yıldızo˘ glu / J. when it is set to 1. Dupou¨ et. When compRange is equal to 1. An agent has a limited memory that does not permit it to deal with the whole problem space by itself. If this distance is higher than compRange. the communication is allowed. This is intended to account for the focused learning taking place in communities of practice. a parameter. When an agent first launches a communication process. the agent considers that the problem is too far from its specialisation area and passes the problem to the next agent in a randomly composed list of agents. of Economic Behavior & Org. Each agent uses an address book to keep in memory successive communications with different partners. it belongs to [0.

The agents only accept to handle problems that are close to their specialisation area. The environment remains the same. In particular. it consults its community (contained in its addressbook). If a member of its community provides it with a satisfactory solution. it will then ask once again consultRatio of the total population of agents. in this new . Hence. it is removed from the address book. 61 (2006) 668–690 675 Fig. they are engaged in a practice involving individual learning. If none of these agents answer its enquiry. these two parts are interrelated in that communication informs practice and practice fosters exchanges. M. which can be seen as an exploration of its social environment and a search for new partners. Behaviors of agents are then twofold. 3. However. However. Solving problems in CP. exval is thus a parameter indicative of the stability of links established by agent and of the general degree of trust in the organization. a counter (exval) is associated with each agent present in a given address book. Yıldızo˘ glu / J. and they are committed to social interactions. an agent engages with other agents in the building and the sharing of a common repertoire of knowledge resources. Dupou¨ et. • in order to enhance its practice. • the agent then nurtures this community of practice with its experience and in turn relies on this community to enhance its practice. of Economic Behavior & Org. If an agent accepts a problem but does not have a solution for this problem.4. this implies that the agent is always willing to cooperate with other potential members of a community and that there is no direct cost associated with the establishment of a new relation. 2. In order to avoid the presence of non-relevant agents in the address book. this solution is integrated in the population of rules of the initial agent (the agent learns this rule). The role of the manager The second organizational structure implemented consists in the simple addition of a manager on top of the structure presented so far. If an agent does not answer to exval consecutive enquiries.O.

organization. then the manager chooses an agent that is not often called and that has a specialisation realm close to the signal submitted by the environment. To choose an agent for answering a given problem. The condition part of the manager’s rules corresponds to the signals arising from the environment. The duty of the manager is to allocate the task of answering a given signal to the agent most suited for this specific problem. M.e. On the other hand. The role of the manager can then be seen as twofold. We have also done experiments allowing the manager to call some (variable) number of agents as long as it cannot find an agent that would accept to treat the problem. several other agents will generally be questioned by the initially chosen agent. each agent is referenced to by an integer). the rule with the best prediction is chosen. From the modelling standpoint. the roulette wheel is used). 4 The fact that the manager only does one random experiment is not really restrictive. it can also be viewed as a guide for the learning of agents. when it is 0. 4). of Economic Behavior & Org. These two values again correspond to a manager who favours exploration (0: roulette wheel) or exploitation (1: best action).676 O. the manager can select the rule with the best prediction of performance in its action set or use a roulette wheel process to chose the action. . The results we obtained are very similar to the ones exposed in the next section. First. 61 (2006) 668–690 Fig.4 In the latter case. The action parts consist of the references to the agents (i. Once an agent has been selected by the manager. Dupou¨ et. 4. the problem is passed to this agent and it tries to answer it. This is controlled by the parameter select Man (when select Man is 1. the manager is an XCS similar to the ones used for modeling agents. On the one hand. since the allocation of tasks made by the manager will influence the learning processes of the agents. Yıldızo˘ glu / J. the manager adds in its memory a new rule of which the condition part is the signal from the environment and whose action part is the reference to the selected agent. it has to allocate efficiently the tasks as they are submitted by the environment. If it does not have any rule. As in the case of agents. When communication is allowed. Problem solving procedure of the manager. the manager proceeds as follow (see Fig. the manager tries to find an agent by selecting in its memory a rule that matches the signal from the environment. the signals from the environment are directly received by the manager.

When the organization faces a problem. for each model. Yıldızo˘ glu / J. As a result. The relative weight of negative performances in comparison with the positive ones nicely indicates the capacity of each organization to treat problems. we adopt a methodology that allows quite a systematic exploration of the parameter space of the model.000 observations covering quite a diversified subset of the parameter space and the problem space. 5 15000 problems correspond to the yearly average number of cases considered by the refinancing company we have analysed. we obtain. We analyze these random samples using box plots (giving the four quartiles of the distributions of the variables). Dupou¨ et. If the organization is a help desk. it can find a solution and obtain the actual performance returned by the environment. for each run we obtain an average number of 150 randomly chosen observations for all the measured variables. The values from which different parameters are drawn can be read in the Appendix A. of Economic Behavior & Org. M. In that case. and incurring a cost that corresponds to a missed opportunity for the organization. the problem remains unsolved (the organization has not been able to develop a policy for this problem). in which case the problem remains unsolved. the agent can either solve it alone or ask for help from other agents (if communication is set to 1). and they can be lost to the competition.u-bordeaux4. The simulation protocol and methodology Given the complexity of the interactions we model and the strong non-linearity of the decision processes of the agents. or it can be unable to provide any Thus. we run 500 series of 15.1000]) is the most direct indicator of the organizational efficiency. projects that are not handled can be warranted by the competition and the firm can loose its credibility and its market share. Missing opportunities can be fatal in a competitive environment as is emphasized by Visser. The performance takes into account this missed opportunity.pdf. The reward is then used by the agent and the manager to carry on their learning processes and refine the parameters associated with their rules. not being able to solve the problems of the customers can be very frustrating for them. Development Core Team. If the organization is the financial warranty providing firm of our example. 3. The gross performance (∈ [−1000.6 We do not necessarily discuss in the text all the parameters that appear in this appendix but only the most significant ones. For each of our two models.O. However. Student tests between subsets. This performance criteria is useful for us in order to distinguish firms that generally can solve problems but miss some opportunities from the ones that are never able to solve problems. a set of 75. 6 This appendix is downloadable from our web site: http://beagle. The selected agent returns its answer directly to the environment that evaluates it and gives a reward. The statistical analysis is conducted using R (see R. the problem is not passed to another agent when the selected agent cannot answer. we must qualify this capability.1. 61 (2006) 668–690 677 To answer the problem. All series are initialized with a randomly drawn vector of values for the main parameters of the model. 2003).000 problems5 each where the results from each problem has a probability of 1% of being saved. Given that we desire to evaluate the capability of the organization to handle problems as well as the quality of this treatment. . Simulation protocol and main results 3. This methodology is close to the Monte-Carlo method. it is computed in this case using the negative of the past average performance (the opportunity that the organization has missed by being unable to solve the problem). histograms and regression trees.

Comparative analysis of the organizational performance We give in Fig. We first proceed by comparing these organizational forms from the point of view of gross performance. Yıldızo˘ glu / J. of Economic Behavior & Org. Comparison of gross performance between organizations. 5. Fig. 5 the comparison of the distributions of gross performance for all cases in both organizational forms.2. If a solution is found after a tremendous amount of communication between agents.1. Dupou¨ et. The Student test results provided below confirm this result and show that the average of gross performance in CP is higher than the one in HD. As will clearly appear in the rest of this analysis. This strong result calls for a more detailed analysis. These boxplots show the four quartiles of these distributions: the statistically significant minimum and the maximum correspond to the extreme end of the whiskers. M. it would correspond to a relatively high organizational cost. the gross performance neglects another important dimension of the efficiency of the organizations: the cost of communication. while Q1 and Q3 correspond to the edges of the central box and the median corresponds to the horizontal line inside the box. These first results are refined afterwards through the analysis of the role of communications and communities of practice. we also analyze the total number of communications in each case. . In order to take into account the cost as well as the benefit of each decision. one of the phenomena that plays a crucial role in the performance of these organizational structures is the possibility of communication. Main results of the simulations We focus the analysis of the results on issues concerning the performance of organizations in problem solving.678 O.2. 3. 6 compares the distribution of the gross performances of HD and of CP between two cases: Fig. The last step of the analysis clarifies the role of different parameters in these results using regression trees. We observe that communities of practice (CP) is globally able to treat more problems than the hierarchy (HD) (hence it better avoids negative performances) and it treats them better than HD since the highest performances have a higher frequency in CP. 61 (2006) 668–690 However. 3.

9367.86. 7.6562. The graphic (a) shows that the communities of practice need communications to better avoid the negative payoffs (due to unsolved problems). (2) Student tests between different cases concerning the gross performance give the following results: H1: CP < HD → t = 64. of Economic Behavior & Org. p − value = 1 H1: CP (0) < HD(0) → t = 55. while the hierarchy needs communications for attaining better payoffs (graphic (b)). p − value = 1 H1: CP (1) < CP (0) → t = 8. M. 6 evidences that com- . p − value = 1 In Fig. The comparison of results between HD and CP shows that CP is able to attain.3433.20. but communities with communication (CP (communication = 1)) attain the highest performances and the complete ordering between these four cases is given by CP(1) ≥ CP(0) ≥ HD(1) ≥ HD(0).6133. The organizational structure plays the main role in the determination of the performance.14. p − value = 1 H1: CP (1) < HD(1) → t = 51. df = 149390. It clearly shows that the number of communications is the highest in CP. p − value = 1 H1: HD(1) < HD(0) → t = 41. Comparison of gross performance between organizations with (1) and without (0) communication. df = 31741. df = 32190.9592. df = 122334. communications do not play the same role in these two organizational structures. with communications (1) and without communications (0).3. df = 12777.2951. p − value = 1 H1: CP (0) < HD(1) → t = 27. Dupou¨ et. df = 18648. better payoffs than HD.O. 6. As a consequence. on average. Yıldızo˘ glu / J. Moreover. 61 (2006) 668–690 679 Fig. we consider the total number of communications in CP and HD. Fig.58.1.

In the rest of the article. having to find a balance between the number of communications that would increase their performance and a reasonable cost of these communications. of Economic Behavior & Org. Cliquishness is the average of this measure over all vertices. Given the assumptions we adopt about the agents and their behaviors. Nonetheless. Wasserman and Faust. in our setting. we will focus the analysis on the settings where communication is allowed.. one can hope to observe the emergence of communities of practice.j) = 1} is the direct neighborhood of i. 2000). such groups would correspond here to communities of practice. 2001. In other words. If G(I.680 O. if this cost is really consequential. Vi = {j|d(i. Organizations are thus facing a trade-off. play a role in the behavior of the system as a whole. the fraction of potential edges that actually exist between them: the full connection of all the n neighbors with each other would correspond to n(n − 1). this cost cannot be specified without introducing an ad hoc bias. Dupou¨ et. 7. In a directed graph. when communications occur. the average cliquishness of G can be defined using the following notation. 3. cliquishness (Newman et al. 61 (2006) 668–690 Fig. However. munications significantly enhance the gross performance of the organizations but they could also be seen as an important source of organizational cost (Ishida and Ohta. we here use an indicator well-known in social network analysis. 1994). In our setting. to allow communication is to allow agents to build relational structures beyond the hierarchy. Watts. outside the hierarchical line.2. It is thus an indicator of the existence of groups of tightly connected agents within the graph. cliquishness represents. Butler. with I being the set of the n nodes of G and d being its adjacency matrix. Emergence and performance of the communities of practice The comparative analysis of performance indicates that communications among agents. d) is a graph. Let ni = #Vi . To account for the emergence of their communities. 1999. which captures the idea that two agents connected to another agent are likely to be connected to one another. If d(i.2. 2001. Yıldızo˘ glu / J. Comparison of the number of communications. for a vertex with n neighbors. M.j) is the distance between the vertices i and j. it could eliminate the contribution of communications to gross performance.

the highest performances are attained mainly in the cases where the communities of practice emerge. These observations will be confirmed below in Figs. are the ones that also correspond to the emergence of the communities of practice (the distributions possess a cumulation around 0. Given that communications in our model create new links. (12a) and (13a). we characterize the emergence of communities of practice in the CP model.3). To do so. We show that full-fledged communities of practice only emerge for certain values of these indicators.3. A potentially interesting feature of communities is their contribution to the attainment of high performances. Yıldızo˘ glu / J.2–0. we observe in the latter either a higher number of communities or more tightly knit ones. In Dupou¨ et et al. Comparison of Cliquishness. of Economic Behavior & Org. the most frequent values of cliquishness.l d (j.l) = 1. As a consequence. degrees and cliquishness). l)j. These results show that for both types of organizations. In order to study the role of communities in the overall performance. but also in HD. 8. Nonetheless. ni (ni − 1) C= (3) i Cliquishness can consequently rank from 0 to 1. 61 (2006) 668–690 681 and mi = j. as well as eliminate the existing ones that become irrelevant. 9 shows that in CP as well as in HD.2 and 0. when the performance is maximal. The cliquishness of G can be computed as 1 n mi . Even if it is practically quite arbitrary to determine a limit value of cliquishness corresponding to the emergence of communities of practice. Fig. we consider the distribution of cliquishness in the cases corresponding to the maximal performance.O. the passage of time alone does not automatically increase cliquishness. Fig. (2003). the graph is made of fully connected subgraphs. When cliquishness equals 1. we use three indicators (betweenness. 8 shows that communities do not emerge solely in the CP model. When it is equal to 0. Fig. . the highest values of cliquishness are only reached in CP. Dupou¨ et. the correspondence between the highest performances and the presence of structures similar to communities of practice is quite strong. l ∈ Vi such as d(j. These configurations correspond to values of cliquishness between 0. M. no triangles exist within the graph.

682 O. Regression trees are very flexible and powerful in the clarification of the structure of the observations. 10. Fig. Dupou¨ et. Yıldızo˘ glu / J. Regression tree for the gross performance in HD. Our results have so far focused on the most striking results of the model. chapter 10). A regression tree establishes a hierarchy between independent variables using their contributions to the overall fit (R2 ) of the regression. 1999. the understanding of the mechanisms behind these results requires further explorations. 61 (2006) 668–690 Fig. Behind the curtains: the role of the main parameters We analyze the role of different parameters of the model using regression trees (Venables and Ripley. 10.3. it splits the set of observations in sub-classes characterized by their value in terms of their contribution to the overall fit and of their predictions for the dependent variables (all parameters that are modified by the Monte Carlo procedure are included as explanatory variables in each of the following regressions). We now consider the determinants of the main performance variables of our model.2. Distribution of cliquishness for the cases with the maximal performance. . We give in the next paragraph a step-by-step interpretation of the regression tree exposed in Fig. This value is validated against a fraction (10%) of the sample that is not used during the estimation. However. M. More exactly. 3. 9. of Economic Behavior & Org.

when the manager chooses the best agent in its agents set for each problem (instead of choosing each agent with a positive probability. 10. compRange and selectAg.2) can only be expected if the agents are not too specialized (comprange ≥ 0.5787) are fulfilled for a set of 20307 observations (over a total of 20307 + 13454 + 5661 + 12875 + 13594 = 65891) in which the expected gross performance is −243. This is a common feature of all trees. This performance is even higher if the agents also favour . The Figs. On the left. (selectMan < 0. corresponding to (selectMan < 0. 3.5). the agents should not be too specialized (compRange ≥ 0.2. In this case.5) in fact corresponds to selectMan = 0. Higher performance is obtained when selectMan = 1 and.O. The condition (selectMan < 0.2.5) and (comprange ≥ 0. For example. the main determinants of the gross performance are selectMan. 61 (2006) 668–690 683 Fig. The other variables play only a marginal role.5787). The final nodes of the tree indicate the number n of observations concerned by these conditions and the expected value of the dependent variable (here the gross performance) in this sample.3. we have 13454 observations verifying (selectMan < 0. The rest of the tree (and all other trees) can be read following the same scheme. positive performances (169. the manager favours exploration over exploitation. Determinants of the gross performance. In Fig. At the other side of the last condition. are again separated in two sub-groups by another condition concerning the specialization of the agents: (comprange < 0. 11.5787). HD with communications In the case of the hierarchy. the first condition (selectMan < 0. of Economic Behavior & Org. the observations that correspond to (selectMan ≥ 0. the ones for which it is false. and on the right. even if it is increasing with the performance.5) and (comprange < 0. we have the observations for which this condition is fulfilled. and on the right. selectMan = 0).5).1.5787). Dupou¨ et. on the left of a condition we have observations for which the condition is true. M. The observations on the left.20) in order to have an interesting global performance. 10 and 11 exhibit the variables that contribute more than 1% to the overall fit in both cases (HD and CP) with communications.5) divides the observations in two subsets. When the manager favours exploitation (it uses the best action rule). hence. Yıldızo˘ glu / J. Regression tree for the gross performance in CP.

Determinants of gross performance in HD. These values are split in two groups. 10]. 11). (a) Role of Cliquishness (b) Role of comprange (c) Role of selectAg. Determinants of gross performance in CP. 12 and 13 give the distribution of gross performance as a function of another variable (for example. Q2 ) and on the right. There is a learning trajectory of the system over time. Even if selectAg plays the main role in the segregation of different cases. the organization is only able to treat problems if the agents are loosely specialized [Fig. on the left we have the values of the performance when the explicative variable is lower than its second quartile (its median. Dupou¨ et. Fig. CP with communications In the communities. When conditions upon compRange and selectAg are fulfilled.13). 61 (2006) 668–690 Fig. The community can only attain significant performances if the agents are not too specialised (compRange ≥ 0. M. exploitation (they use the best action rule (selectAg = 1) for choosing their actions). When the manager uses a more explorative strategy (selectMan = 0). the hierarchy between the determinants is slightly different. Yıldızo˘ glu / J. we have the complementary cases where the variable is higher than its median. (a) Role of Cliquishness (b) Role of comprange (c) Role of the behavioral selectors. of Economic Behavior & Org. Their role only appears when we consider more qualitative aspects of the performance (Fig.684 O. then time matters (since the variable period correspond to the number of the submitted problems). 12. it seems that the initial conditions in CP are less determining than in HD. . The boxplots of Figs. It is interesting to observe that other dimensions of the communication (like the consultRatio) play only a marginal role in the determination of the gross performance. Since time does not play this privileged role in HD. 13. the specialisation of the agents clearly appears as the main determinant of the performances. cliquishness).

O. HD with communications In HD the number of communications remains relatively limited. but this is not enough. Dupou¨ et. The number of communications in the organization will generally imply a cost as we have stressed above (Ishida and Ohta. M. Determinants of the number of communications. Even if strongly restricting the possibility of consultation is not possible.3. This is not a tautology. It is legitimate then to ask if it is possible to limit the communications when their cost is prohibitive. Butler. Yıldızo˘ glu / J. 3.17. etc. an intermediate level of specialisation (compRange ∈ [0. while it increases the accuracy of problem solving by HD.63]) of the agents can also be used to contain these costs (Fig. CP with communications In CP. . Whatever the nature of this cost (linear. We should not forget that even if the manager and the agents favour exploitation. some minimal amount of exploration is always present through the genetic algorithm. 2000). 14 in order to identify the parameters that the organization could weight in order to reduce this cost. 2001. Some limitation of the possibilities of consulting other agents is necessary Fig. Last but not least. 12 with Fig. 13 reveals an interesting structural regularity: higher performance is associated with higher cliquishness and lower specialisation (higher comprange).2.0. since the actual number of communications also depends on other variables such as the specialization of the agents. This again confirms the positive relationship between the emergence of the communities of practice and the global performance and explains how this mechanism works: higher cliquishness increases the capacity of CP to cover the problem space. We consider the determinants of the communications in Fig. of Economic Behavior & Org. and exploitation allows the hierarchy to increase its performance while permitting the communities to reduce negative performances. Determinants of communications.) it will generally be increasing with the number of communications. 14a). a manager who strongly favours exploitation over exploration in its decisions will also be able to limit these costs. 61 (2006) 668–690 685 The comparison of Fig.1) for the agents. It is sufficient to limit the possibility for the agents to consult other members in order to minimize these costs. 14. non-linear. one must definitely avoid a very high specialisation (compRange < 0.2.

Figs.) and above.2. especially in CP (lower compRange corresponds to a considerably higher number of communications). We find here the trade-off emphasized by Bolton and Dewatripoint between specialisation and the cost of the communications. Moreover.43). .1) with a reasonable weak consultRatio (<0.3. costs can be contained if agents are incited to favour exploitation (selectAg = 1) (Fig. Dupou¨ et.3 in HD and not too specialized agents in CP (compRange > 0. given that consultRatio does not strongly condition the gross performance (it does not appear in Figs. given that we take into account the quality of the treatment of the problems by the organization (a dimension neglected in Bolton and Dewatripoint) the problem of the organization is truly a trade-off. HD can successfully assure such a trade-off. 10 and 11 and the ones of Fig. 10 and 11). 61 (2006) 668–690 to limit the communication costs as in HD (consultratio < 0. When this is again not possible. In particular. Determinants of cliquishness. 15 and 16 show that the determinants of the social structure are slightly different in these two organizational structures. 15. On the other hand. of Economic Behavior & Org. 14b). since the limitation of the communications will generally have a negative impact on gross performance. Fig. it is possible to overcome this trade-off: the highest performances are attainable even if we have consultRatio < 0.3. M. The role of cliquishness in the characterization of the emergence of communities of practice has been studied by us in a previous article (see Dupou¨ et et al. Regression tree for cliquishness in HD.43). but it can also incur a tremendously higher level of communications. As a consequence. 14. CP can assure more systematically positive performances.3. We also remark that without such fine-tuning.686 O. Yıldızo˘ glu / J. 3. given the very low level of communications in this structure and the reasonable level of the performances. we have established in this article that the communities of practice should correspond to a cliquishness around 0. the global efficiency can be reinforced by crossing the prescriptions of Figs.

When 0. M. that is. In any case. 15).5 implies a cliquishness of 0.5 to obtain the highest values of cliquishness. exval still has to be above 3. Hence. HD with communications Three variables intervene in the emergence of communities in the case HD: consultRatio. in order to build communities of practice. but . when consult Ratio is low.e.11 ≤ consultRatio < 0. agents need both a great ability to screen their social environment and a certain stability in their relationships (Fig. CP with communications The same three parameters are the determinants in the emergence of communities in the case CP. Yıldızo˘ glu / J. exval ≥ 9. there must be some kind of fidelity or trust in relationships between agents in order to observe the emergence of communities. of Economic Behavior & Org.11). 61 (2006) 668–690 687 Fig.35. compRange (that is the level of specialisation of agents) plays here a less important role than in the case HD. Even when consultRatio is higher. As in the previous case. compRange has to be above a certain threshold to have values of cliquishness corresponding to full-fledged communities of practice (Fig. Dupou¨ et. Regression tree for cliquishness in CP. However. This is of course not a very surprising result in itself. then exval has to be quite high. 16. then no communities can emerge and the value of cliquishness is almost one of a random graph. The main determinant of the importance of communities of practice is therefore the possibility of communications between agents. When consultRatio is low (consultRatio ≤ 0. The highest values of cliquishness are attained for high values of consultRatio and exval. 16). high values of exval help reaching reasonably high values of cliquishness (i.O. exval and compRange.14).

CP can assure more systematically positive performances. for example). 61 (2006) 668–690 given that the conditions for it are quite demanding (consultRatio > 0. The particular role that such agents could play in the network can be very interesting to study. the existence of an intranet.35). we have assumed that the agents are always willing to enhance their individual competencies by resorting to communities. and communities of practice are not a simple epiphenomenon. M. In these cases. Conclusion This article analyzes the potential impact that informal social structures. particularly if one considers learning in the long term (i. Moreover.). We should not forget that the structure of the communication hierarchy is endogenous in our model and its structure could be more directly compared with the structures considered by Bolton and Dewatripoint (conveyor belt or strict hierarchy. Dupou¨ et. When the reliability of the performance is the issue. This paper thus backs the claim made by Bowles and Gintis (2002) that hierarchy and communities are complementary rather than substitute modes of governance. communities of practice. However. Regarding communities. It is the interplay between agents’ idiosyncratic capabilities and manager’s behavior that ends up in an organization’s specific outcome. some indicators from the social network analysis literature could be used to characterize the hierarchical structure of the agents’ communication network. some of them being less specialized than others. without strong time constraints). the examination of the interrelations between variables reveals that two of them are especially crucial for the emergence of communities and their performances: the degree of specialisation of agents and the conditions of communication between them. When the cost of communications are important. some degree of trust in the organization is also necessary. the detailed dynamics of their behavior has not been explored.g. Yıldızo˘ glu / J. This assumption . Also. can have upon the performance of firms or subunits of firms. 4. In the case of HD. The first result of this work is to show that community structures are efficient for competence building. we are well aware that this can only constitute an exploratory work on these issues. etc. an interesting development of the model could follow Hart and Moore and allow for a differentiated specialization of the agents. However. the efforts made by the management to ease communication. In particular. one must add the behavior of management to the two previous parameters. the legitimate peripheral participation process that Lave and Wenger put at the heart of the evolution of communities’ structure has not been explored. this work tends to show that hierarchy altogether with communities of practice is the most efficient structure among those we explore. it implies a very low level of communications and a reasonable level of performance. but it can also incur a tremendously higher level of communications. Besides. HD can successfully assure a satisfactory trade-off. the second can be seen as part of the environment in which communities exist (e. of Economic Behavior & Org.688 O. communications are not enough. The way managers apprehend their task has an influence both on the conditions of emergence of communities within the firm and on the global performance of the organization. real firms are required to answer demands from the market in a timely fashion. In an article more directly focused on the hierarchy (instead of the communities of practice).e. concerning the incentives and motivations of agents to enter in a community. Enriching the possible structures for the hierarchy could also be very interesting since this would allow a better comparison of our model with the hierarchy models developed by Bolton and Dewatripoint or by Garicano. If the former clearly depends on the individual agent.

Advances in Learning Classifier Systems.. The Journal of Political Economy 108. belief development and structural change: a dynamic model of networked communities of practice. M. Moreover. 119–137. M. Z. International Journal of Industrial Organization 20. Financial support of the CCRRDT program of R´ egion Aquitaine is gratefully acknowledged by Murat Yildizo˘ glu. (Eds. Friedberg.cmu. Duguid. P. Paris.. Berlin. Dupou¨ et. Induction. M. Management Science 43. 2003.. L’acteur et le syst´ eme. An algorithmic description of XCS. B. Cambridge. S. Points Seuil. Brown. 1977. K. Butler. For instance. 40–57. 2003.. Organization Science 2.. whereas this is certainly a main concern in management studies today.. M. J.. Moore. 2005...R.V. London..W. 675–702. California Management Review 40. All these limitations highlight the work that remains to be done in modeling the behavior and the role of communities within firms. Revue d’Economie Industrielle 103. 419–437.). Dewatripoint.. O. the fostering of communities of practice is not considered as a strategic issue by the manager. P. W. The Economic Journal 112. We are very grateful to two anonymous referees who have considerably contributed to the readability of this article. . P. The firm as a communication network. Oxford University Press. Lecture Notes in Artificial Intelligence (LNAI-1996). Communication cost. Processes of inference. The determinants of the management hierarchy: evidence from italian plants. H. 2001.. E. Springer-Verlag. 976–997.. 1994.S. Holyoak. Organizing knowledge. In: Lanzi. Garicano. Gintis.. A. 2000. J.O. Thagard.W. Yıldızo˘ glu / J.S.. J. K. Third International Workshop (IWLCS-2000). Journal of Political Economy 113. 91–110. in Carley and Lin). S. Wilson. Lin. Hart. It could also be interesting to investigate a more rigid hierarchy where the manager strongly prescribes the tasks to agents without seeking to take into account their learning curves or specialisation (as. Butz. Social capital and community governance.. MIT Press. M. Our model with hierarchy also focuses on some key aspects and leaves many other features unaddressed. Brown. 874–904.. Wilson. Stolzmann.. O. 809–839. 90–111. 1998.. Acknowledgements We gratefully acknowledge the comments and suggestions of the participants of the SCE Workshop on Complex Behavior in Economics and Finance 2003 and of the WEHIA 2003 Workshop. 1997.. 2003) should also be analyzed since they can condition the learning process of the agents. Crozier. P. 61 (2006) 668–690 689 is common in the literature about communities of practice but would certainly deserve a deeper exploration. 2002.casos.H. 1991. Carley. M. P. The Quarterly Journal of Economics 109.M. 2000.. References Amin. 2002.. L. Bowles. Morphogen´ ese de communautes de pratique.L. we here only dealt with a specific form of hierarchy where the manager is engaged in a learning process. Hierarchies and the organization of knowledge in production. CASOS working paper. S. Delmastro. Knowledge Practices. Cohendet. Yildizo˘ glu.. More complex problem spaces (as in Rivkin and Siggelkow. Duguid. http://www. Dupou¨ et. learning and discovery.. 1989. Cohendet. It is hoped that this work contributes in opening the way to this future research. Bolton.J... On the design of hierarchies: coordination versus specialization. P.ece. of Economic Behavior & Org. learning and innovation. Organizational learning and communities of practice: toward a unified view of working.. P.S. Holland. A theoretical study of organizational performance under information distortion. J... Communities and Competences in Firms. for

Wenger. E. Vienna: http://www. P. of Economic Behavior & Org. 149–175. Social network analysis: methods and applications. Lave. Small worlds. 2000. 2001. Springer-Verlag. Princeton University Press.. S. Berlin. Venables. Classifier fitness based on accuracy. 1994. 2000.J. 263–283.. R: a language and environment for statistical computing.r-project.. Computational & Mathematical Organization Theory 1 (1). Siggelkow.W. Cambridge University Press.. 33–63. 3rd ed.D. S. Vyhmeister. Cambridge University Press. Communities of practice: learning. 2000. (Eds. B. Evolutionary Computation 3. Cornell University K. N. Watts.L. R. Rivkin. meaning and identity. 1990.L.... Dupou¨ et. Springer. P.W.. T.. Cambridge.. Organizational communication structure and performance... J. Yıldızo˘ glu / J. New York. T. Hogg.C. 231–252. Learning classifier system. Riolo.J.W. 1999.. Wilson. In: Lanzi. 2000. NY. Ishida. 026118. 290–312. K. Wilson.N.. 311–330.. Ohta. R. S. Performance of organizational design models and their impact on organization learning. Communities of practice: performance and evolution.. Strogatz. D. 1998. B. from foundation to application. Wilson. Computational & Mathematical Organization Theory 7. Management Science 49. 2003.. S. W.). W. In: Lanzi. B. 1995. . Faust.. S. M. pp. An adaptive agent based economic model. Ripley. Ross. Talking about machines: An ethnography of a modern job.690 O. (Eds. Watts. M. Stolzman. Wasserman.. 1999.. Lanzi. Berlin.. Visser.. New York. Wenger. A. Orr.). Springer-Verlag. Balancing search and stability: interdependencies among elements of organizational design. Cambridge. Modern applied statistics with S-PLUS. Random graphs with arbitrary degree distributions and their applications. Development Core Team. Computational & Mathematical Organization Theory 6. J. On a mathematical comparison between hierarchy and network with a classification of coordination structures.A. from foundation to application. S.. Stolzman. J.J. 61 (2006) 668–690 Huberman. 1990... E.. Journal of Economic Behavior and Organization 42... Situated learning: legitimate peripheral participation. Schulenburg. Cambridge University Press. 2001. 2003. R foundation for statistical computing. pp.L. 395–410. 1995. Ouksel.E. A roadmap to the last decade of learning classifier system research. P. Physical Review E 64. W.H. 73–92. Ithaca.W. P.. D.. Newman. R. Learning classifier system.