The Agent-Based Model

ABM is a way to model the dynamics of complex systems and complex adaptive systems and offers many significant advantages in this regard. For instance, ABM overcomes potential modeling restrictions that occur in a system of equations due to collective behavior, knowledge interactions and learning loops (Parunak et al. 1998, p. 21). Two further points are emphasized by Pyka and Grebel (2006): First, the possibility to exhibit how collective phenomena arise through interactions of autonomous and heterogeneous agents and second, the possibility to analyze various institutional arrangements and different potential paths of development in their particular decision contexts (Pyka and Grebel 2006, p. 24).

According to Macal and North (2010, p. 152) three key characteristics are inherent to ABMs:

  • 1. A set of agents, their attributes and behaviors
  • 2. A set of relationships and methods of interaction
  • 3. The agents’ environment

The used agent-based model of the Vienna life sciences is a simplified and tailored version of the agent-based simulation model (ABM) of the Vienna live sciences innovation system developed by Korber (2012). The model by Korber itself is based on the SKIN model (Simulating Knowledge dynamics in Innovation

Networks), an ABM containing heterogeneous agents which act in a complex and changing environment (Ahrweiler et al. 2004, p. 285), and relies on previous research by Gilbert et al. (2001, 2007), Pyka et al. (2002) and Ahrweiler et al. (2004).

The present simulation model used in this chapter contains multiple empirically based autonomous agents, which are heterogeneous with respect to their attributes. There are three agent types: university agents (including universities of applied sciences), research organizations agents (public or private non-profit research organizations) and industry agents (small and medium-sized enterprises including start-up and spin-off companies and large enterprises). Each agent is characterized by one or more research fields, an associated particular core competency and a corresponding expertise level. The agent type, the agents’ attributes, research fields and core competencies are empirically founded, in a way that each agent in the simulation model corresponds to a real-world firm or researching entity of the life sciences sector in the Vienna region.

The research orientation of the agents may be either no research, basic research or applied research, which is an empirically based attribute in the simulation model. Additionally, agents may differ in their research attitude (i.e. incremental or radical partner choice for collaboration), research strategy (i.e. whether to conduct own research or do research in collaborations), partner search strategy and collaborative strategy. Further attributes are amongst others the financial stock, number of employees, number of researchers and foundation year (for a table of agents’ attributes see Table 3).

The attributes research field, core competency and expertise level are combined to characterize the knowledge endowment Ki of an agent ai where i = 1,2, ..., I. Each Ki consists of a set of kenes ki (Gilbert 1997):

Hence, the agent’s specific knowledge endowment is given by:

where R denotes the set of all research fields rm : R = |rm|m = 1, 2, ..., 35}, C the set of all core competencies cn : C = {cn|n = 1,2..., 6}, rm the research field m, cn the core competency n and yimn the expertise level that agent ai has in research field rm and core competency cn; {yimn 2 N|0 < yimn < 10}. The Vienna life sciences sector can be divided into 35 research fields subsuming scientific or technical fields (see Table 2 in the appendix for the list of research fields) and six business domains (i.e. R&D, Contract research, Sales, Service and Education/ training) represented by six core competencies in the model (Austrian Life Science Directory 2012).

With regards to the behaviors, each agent decides at the beginning of the simulation about its research strategy, i.e. whether to conduct exclusively own research (go-it-alone) or do research in collaboration (Pyka et al. 2002, p. 176; Ahrweiler et al. 2004, pp. 6-7). Once the agent decided to do research cooperatively, it has the possibility to constantly search for collaboration partners and not doing any research on its own (imitative strategy) or do own research as well as in collaboration with other agents (collective strategy). Obviously, potential partners have to be found in order to perform cooperative research. This can be done either conservatively, where agents with similar research fields are more preferred, or progressively, where agents with less research fields in common are favored (Ahrweiler et al. 2004, p. 8). It has to be emphasized that only agents with the research orientation basic research or applied research may perform research.

For every agent, a research concept Di is formed using its knowledge characteristics. Therefore, three kene triples are randomly chosen (the probability is proportional to the expertise level) from the agent’s knowledge endowment Ki. Hence Di is given by:

where ^ k"i, k 2 Ki as weH as kli = (C ^ yfimn), k = (rm, ^, yL) and

ki (rm; cn ; Уimn ) .

Thereafter, the research concept is evaluated to determine if the research concept was successfully used for the creation of inventions. This is done via a fitness function, where the sum of the expertise levels yimn, Y"mn and /-mn in the research concept Di is compared to the average sum of the expertise levels of the other agents’ research concepts (y'jmn, yjmn andy'jmn). If the sum of the expertise levels in the research concept of agent ai is equal or above the average of the other agents, the research concept is considered to be successful.

An indicator variable is generated for the output generation:

where fi = Yimn + Yimn + Yimn and fu = Y'jmn + Yjmn + fjmn.

If ft = 1, the research concept D" is successful and transformed into either products (i.e. commercialization on the market), scientific publications or patent applications based on the application orientation O" of the agent a".

Consequently, if the research concept was successful and o; = 1 (i.e. no research), the output is measured in terms of commercialization on the market. Similarly, if o; = 2 (i.e. basic research) and o; = 3 (i.e. applied research) the output is measured in terms of scientific publications as well as patent applications.

Agents who perform research (i.e. with research orientation basic or applied research) have the ability to learn and also to forget. Learning may occur through learning by doing, or learning by interacting (Pyka et al. 2002, p. 174). The expertise level is increased by one if the kene triples are part of the research concept and hence are used for current research. In addition, agents may also acquire new research fields. Which research fields are chosen is specified by the agent’s research attitude. According to the thematic proximity of research fields (measured by the Jaccard-index),5 the agent—following an incremental strategy—opts for the most similar research field. Following the radical strategy, the most distant research field is favored (Korber and Paier 2011, pp. 607-608). The choice of the (to the new research field) corresponding new core competency depends on the core competencies already held by the agent. Also, the new expertise level is set to one (beginner-level). The research costs diminish the financial stock of the agents. However, agents may also forget. If the kene triples are not used, the expertise level decreases. If the expertise level in every kene of an agent’s kene set drops to zero, the agent is culled due to poor performance. Also, if the financial stock of an agent gets smaller or equal to zero, the agent exits the system.

Relationships between the agents are mainly implemented as knowledge interactions. These subsume collaborative research, extra-regional relations and the creation of start-ups and spin-offs (Ahrweiler et al. 2004, p. 9). Collaborative research can either take the form of partnerships or networks. The set of potential partners for partnerships subsumes those agents, whose research strategy is collaborative and who are engaged in basic or applied research. From this set of potential partners, a randomly chosen discrete number of partners are asked to form a partnership or network, respectively. The minimum of partners is set to one; the maximum number of partners is four. Agents learn through the interaction in partnerships and create cooperative research results. Only the knowledge that is used to create output increases an agent’s knowledge base. That is, only those kenes that are actively used in the cooperation for the creation of research results can be taken over from other members. If the partner agent is active in research fields that are not already in the kene set of the agent, kenes are taken over from the partner

5The Jaccard-index (Leydesdorff 2008, p. 79) is defined as Jr. r. 512 , where rj and r2

2 Xr1 +Лг2 Xr1r2

denote research fields with r1,r2 2 R = {rmm = 1, Mg, Xr1r2 denotes the number of

co-occurrences of research fields in organizations and Xr1 and Xr2 denote the occurrence of research field r1and r2, respectively.

with the new expertise level set to one and the new core competency set to the most frequent one of the already existing kenes. For those research fields that are already in the agent’s kene set, the expertise level is set to the value of the highest expertise level of the collaboration partners. The cooperative research concept is randomly chosen from the joint kene set of all members. Extra-regional relations are based on the fact that many organizations are active in EU projects. This is accounted for in the simulation model by granting 87 % of randomly chosen agents a financial support (Heller-Schuh and Paier 2009, p. 162). Moreover, the agents participating in extra-regional relations gain an additional random kene for their kene set, indicating extra-regional knowledge input. If an agent was successful (i.e. the sum of the expertise levels in its research concept is above the average of the rest of the agent population), with a certain probability, it creates a new agent in the form of a start-up or a spin-off organization. The new organization adopts the research concept of the successful agent, but with an expertise level equal to two. The financial stock of the newly created agent is dependent on its agent type.

In summary, the basic structure of the simulation model can be divided into an input side, an interaction part and an output side. Agents receive financial resources such as public (direct, indirect and institutional) and private funds and are provided with a certain knowledge base, defined as the sum of the agent’s kenes (during the initialization). In a next step, according to their specific knowledge endowments, they are able to engage in knowledge interactions. The resulting output is then measured in terms of numbers of patents applications, scientific publications and the number of created high-tech jobs. A detailed overview of the model structure and its formalization is given in Fig. 1.

Formalization of the model

Fig. 1 Formalization of the model

 
Source
< Prev   CONTENTS   Source   Next >