INTRODUCTION

Located in northwestern Mexico, and in spite of being the cradle of the “Green Revolution”, the State of Sonora is delayed in terms of biotechnological innovation.³ However, in such region there is a vibrant academic community (researchers, graduate and undergraduate students and at least three institutions and research centers) working on the biotechnological sector. These people are a solid base of specialized human capital that suggest the possibility that this network enquires on emerging phenomena.

The weakness of this nascent network is noticed both in the scarce presence of startups in this sphere and poor patent generation, despite the existence of a large number of research projects from institutions and research centers that cannot become a firm.

Such institutions, which come from the agricultural, livestock and fishing boom in the state, among them Universidad de Sonora, Instituto Tecnológico de Sonora (ITSON) and Centro de Investigación en Alimentación y Desarrollo (CIAD), gather more than 400 professors ascribed to the National System of Researchers and more than 1000 graduate and undergraduate students. For example, Universidad de Sonora is part of scientific communities of maximum level in five or more fields (material science, agricultural sciences, physics, geoscience, engineering, mathematics and chemistry). Publications concentrate in physics (267), engineering (147) and chemistry (218), according to data from Web of Science for 1981-2003.

The universe of institutions and highly specialized scholars in the field of biotechnology is a valuable quality to potentiate the development of such activity, whose potential justifies studying the interaction as a network of students, entrepreneurs, scholars and researchers, their environment and the possible implications in the development of this emerging sector.

Such interactions are a black box: a system that involves a promising dataset of entries and another, rather discouraging of exits (few firms or startups, patents, utility models and projects, though many research articles); we are not aware of the intermediate relations and processes.

Biotechnology is of vital importance for Sonora, not only because it is a semi-desert zone with great challenges as regards soils, risks and weather conditions, but also because these conditions may become opportunity areas from biotechnology, which make part of the frontier research at present (^{Kafarski, 2012}; ^{Matyushenko, Sviatukha & Grigorova-Berenda, 2016}).

The aim of this work is to explore the dynamics of interactions recorded in the biotechnological community of the State of Sonora; to do so, an agent-based model, inspired by SNA (social network analysis) and CAS (complex adaptive systems), proposed by ^{Beckenbach, Briegel & Daskalakis (2009)}, was partially applied to identify innovation networks and signs of emergence of regional economic agents. It is intended to verify the convenience of these tools to study the behavior of key variables in a RIS (regional innovation system) in formation.

Complex adaptive systems are an interdisciplinary research field that seeks to explain how a large number of simple entities organize, without the mediation of a central control, becoming a collective whole that creates patterns, uses information and in some others, evolves and learns, and it is increasingly resorted to understand any kind of system. The study of CAS focuses on a system’s complex, emerging and macroscopic properties. For its part, the SNA theory provides us with the necessary tools to 1) create networks, 2) obtain measurements, and 3) detect clusters.

Such tools facilitate the exploration of the aforementioned black box, and also test our hypothesis, in which we consider that agents in the region’s biotechnological sector operate separately, with scarce scientific collaboration and poor orientation toward innovative practices, which negatively affects RIS efficiency.

The fundamental nature of replicating empirical evidence to build scientific knowledge also motivates this work, as much as the need to rebuild, partially at least, the algorithmic procedures of the referred study, which are not publicly available.

The text is organized into four sections. The first one reviews the fundamental theoretical concepts of emergence and complexity in RIS, and the importance of the network analysis approach and the agent-based models. Following, the creation of networks in the regional biotech sector is discussed, from which a set of data is obtained and turned into an agent- based model. This model is supported on the characterization and segmentation of agents according to a series of categories, whose definition comes from Ajzen-Carnegie’s theory of planned behavior.

In the fourth section, the application of the model and the results obtained are approached, as well their implications for the Sonora’s RIS biotechnological sector are considered; all this in an environment characterized by an accelerated conversation of which it is essential to partake. Finally, general conclusions are put forward and recommendations are made as regards enhancing and enriching research based on this sort of models.

EMERGENCE AND COMPLEXITY IN REGIONAL INNOVATION SYSTEMS

RIS function as a sort of complex interaction network between nodes that operate as agents of interchange of knowledge and various goods, part of a series of systems (cultural, economic, etc.) that work under their own logic, this is to say, they respond as a CAS; it can receive several degrees of incentives that unleash various and unpredictable responses. The Complexity Theory indicates that a high degree of stimulation produces a range of activity close to chaos; in such range emergence occurs. Excessive stimulation leads to chaos, while very little produces equilibrium or inertia.

In CAS everything is in permanent change; inside them, patterns always depend on the context (^{Taylor, 2003}). The main difference between a complex adaptive system and another that is not is the nonlinearity derived from the simple addition of the parts, as it is the case, for instance, of the global financial system (crises, recessions) or the immune system (fever, antibodies).

ER INNOVATION AND SCALE-FREE NETWORKS IN SONORA ’S BIOTECHNOLOGICAL SECTOR

The difference between ER (Erdös and Renyi) networks and Scale-free Networks is their topology; whereas ER networks are designed after an associational random variable –or several –, Scale-free networks exist on their own (^{Hein, Schwind, & König, 2006}). An example of this is the hyperlinks' network that make up internet, or a network of mentions within a micro-blogging system.

In ER networks, clustering coefficients produce a normal curve, this is, there is little deviation or all the nodes are somewhat connected within standards, which does not occur in Scale-free networks, in which a few nodes tend to be closely connected, which produces a peak and a “fat tail” of many nodes scarcely connected. The difference is defined by the way in which both sorts of networks are built. On the one side, in an ER network the addition of new nodes is rather uniform among the existing nodes; while on the other, in Scale-free networks new nodes most frequently adhere to highly connected networks (^{Hein et al., 2006}).

Another difference is the influence that one or the other topography has on the process of information dissemination. While in the ER model there are statistic thresholds under which it stops or does not affect all the nodes, the result on Scale-free networks is utterly different. In this case, technically there is no threshold, this way all the nodes may be affected: tragic if it is an epidemic, though benefic for knowledge generation.

Scale-free networks are a closer representation of networks than traditional ER networks, as they are in “the real world”; however, they are more difficult to translate into simulation models, especially if nodes are individuals.

APPLICATION OF AN AGENT-BASED MODEL ON A RIS

The network previously described only represents a time slot, i.e., a manner of a photograph of the temporary space over which this research was carried out. Even if it keeps information on 1) the topology of the network created by the agents, 2) the agent's attributes at the level of firm, and 3) the multiple contents that demarcate each link, it does not allow us to deepen into its morphology, and consequentially, in the multiple feedbacks between the agents' states and the network as a whole (^{Beckenbach et al., 2009}).

The only way to enter such “new world”, in which the network behaves dynamically, is using a simulation model, in which the agents' attributes or empirical information are turned into variables, which as a whole define a state of the firm. This way, agents are capable of reacting after a stimulus, feedback or threshold set by the application process of a model, evolving over each cycle or iteration.

To illustrate the application process of an agent-based model, we resort to the model proposed by ^{Beckenbach et al. (2009)}, which was applied at a large scale in Germany. We believe that it is a useful approximation in emerging regions, for although the necessary information to apply standardized innovation surveys is missing, it enables deepening into relevant aspects for innovation processes such as leaning toward scientific cooperation and trust in regional partners.

^{Beckenbach, Briegel & Daskalakis (2007)} identify four behavioral aspects relevant for innovation processes: 1) declarative and procedural knowledge; 2) abilities (in terms of finding new heuristics and recombination and association capacities of the elements of knowledge); 3) intrinsic and extrinsic motivation; and finally, 4) personality traits (curiosity, risk acceptance, et cetera).

The interaction between the innovation process' social, competitive and individual dimensions may be characterized establishing a typology of agents and another of competence. For example, one type of agent may be characterized with a combination of pessimistic expectations, innovation goals of reactive nature and certain loyalty to paradigms, while a type of competence, for instance, may be implying low concentration degree, low entrance barriers and orientation to the dimension of cost (^{Daskalakis, 2016}).

APPLICATION OF THE MODEL AND RESULTS

The implementation of the model was carried out on the basis of eight agents using variables that measure attitudes (risk aversion, curiosity, goals) and regulations (intended profit, obtained profit), which motivate the election of one or another mode of action, as well as intellectual and economic endowment variables of the firm (specialized knowledge, financial capacity). The criterion to choose an initial mode of action is modulated by three behavioral parameters (aspirational attitude β, imitation propensity and cooperation propensity).

Following the characteristics of the model are described :

If we name

g – incomes obtained in the last cycle or iteration

s – corresponding aspiration level

The mode of action is defined by the following criteria:

We have that the election of the mode of action works as a function of income rates / aspiration level. Therefore, if the firm aspires to earn 100 pesos over the quarter, and in the last one it earned 200, i.e., twice as much, the rate is higher than the aspirational attitude, thus the mode of action routine, or keeping things as they are, is chosen. If, on the contrary, such rate is below the aspirational attitude –this is, the firm obtains revenues under its expectations –, but higher than the result of subtracting the value assigned to variable i (imitation propensity)⁷ from aspirational attitude , imitation is decided as a mode of action . For example, if the firm intended to earn 100, but it only obtained 50, imitation propensity is high; there is motivation to come up with a new product, even though it is only a copy of other items in the market. In the case of the modes of cooperative and individual innovation, the variable x (cooperation propensity) has to be included.

The above is carried out by means of the code:

ask turtles [

if g / s >= B [set color blue ] ;; routine

if g / s < B and g / s >= B - i [set color pink] ;; imitation, i is imitation propensity

if g / s < B - i and g / s >= B - X [set color yellow]

;; cooperative innovation, X is cooperation propensity

if g / s < B - X [set color red] ;; individual innovation

forward 3

At this point, by running the code above we obtain the total number of agents (turtles) on screen, each one identified with a color that indicates the mode of action chosen according to the parameters defined through the research questionnaire.

Unlike aspirational attitude β, the aspirational level updates at the end of each step, following the equation :

where Φ represents each firm's adaption flexibility. In NetLogo, the previous expression is written as follows :

set s1 ((1 - fa) * s) + (fa * g)

set s s1

Hence, using these expressions and the available dataset, we notice that in the case of the firms under study in Sonora's biotechnological sector, initially most of them locates within the cooperative innovation mode of action. By the 20^th iteration, 50 percent of the firms moves to routine, while the other 50 percent remains in cooperative innovation mode. By the 30^th iteration, all the firms operate in routine mode; hereafter, this behavior repeats.

Variables i and x, which represent each firm's imitation propensity and cooperation propensity, respectively, are defined by the answers to a set of questions associated to such phenomena in the research questionnaire.

Using the notation by ^{Beckenbach et al. (2009)}, the innovation network's TC (Transaction Cost) is a function of knowledge specificity (s), the number of knowledge components (q) and trust (r) in terms of successful knowledge transferences in the past.

F2 (Innovative Force) is defined by:

Where: α represents risk acceptance; cin, the expected cost of innovation; and, f(i), curiosity or inclination to explore with a value of 0; intended income with a value of 1; and, intended market share with a value of 2.

Moreover, F1 (Imitative Force) is defined by:

And F0 (Preservation or Routine Force) is 1.

to obtain F2, it is necessary to solve

Where: w0 represents proclivity to explore; kr, knowledge reserves; fr, financial reserves; asp, the level of intended incomes; and asm, the level of the intended market share. E1 and E2, market elasticity parameters, are used with values of 8 and 16 in the reference standard parametric constellation utilized by ^{Beckenbach et al. (2009)}.

To derive relevant measures, it is necessary to distinguish various sorts of behaviors firms; as regards the model, three sorts are classified: radical innovators, imitators and firms that operate following routines. In the case of Sonora, the distribution of the sort of firms is as follows: 80 percent locates in category F_IR (innova tors) ; while 20 percent does in the category of imitators (F_IIM). None of the respondents identified themselves as a firm with no innovation (F_ROUT).

As it is an emerging sector, there is a reduced number of firms in the region, most of them recently created and running on public / private funds obtained from calls, awards and innovation incentives. All the surveyed firms have a product in the market.

After identifying relevant parameters, two are distinguished:

Given the importance of innovative cooperation for RIS, it is necessary to explain the microeconomic conditions for this sort of connected activity in the model's context. The triggering conditions for innovative cooperation come from the union of behavior and empirical observations (^{Beckenbach et al., 2009}).

The authors distinguish the combination of personal attitudes, subjective norms and conditions that influence the orientation of an agent’s will toward a cooperative mode of innovation. At first, the three forces that forge innovation capacity are basic for the agent’s willingness to cooperate in terms of innovation; these forces are related to agent’s current market performance of the agent. Secondly, the various sorts of agents have different cooperation propensities. This considers that not only is the current market position important, but also some deeper attitudes that come from several communication styles in a number of innovation horizons. Third, at least in a regional context, the frequency of cooperative innovations may be observed. This may help reduce the uncertainty associated with this sort of innovative activity, as it simplifies searching for partners and demonstrate possibilities to overcome opportunism. Fourth, subsidies from political institutions as an exogenous incentive to produce cooperation.

The results of ^{Beckenbach et al. (2009)} are virtually impossible to replicate with a small set of data and a RIS in formation. However, implementing the firms' election of the modes of action and the variation in their behavior over the iterations shows in which way it is possible to codify these equations and translate them into a programming language that allows analyzing all sorts of data.

In the absence of metrics, it is possible to resort to other strategies, as in fact it is made in the model described. Some of them are the use of arbitrary constants and/or parameters defined randomly within certain limits.

In the case of our model, we decided to use a Boolean variable to find out whether a firm possessed differenced knowledge, then going on to establish a cooperation link or a match each time the iteration finds two firms whose variable value is 1, and this link has not been previously established.

The code is the following:

set asset-specificity? one-of [false true]

show [asset-specificity?] of turtle 0

show who

show match

if asset-specificity? = true [set potential-match potential-match + 1]

if counti = 0 and asset-specificity? = true [set match who

set counti counti + 1 ]

if who != match and asset-specificity? = true and potential-match > 1 [

create-link-with turtle match

set potential-match 0 set counti 0]

fd 1

show count links

The code above produces the next exit after eight iterations, cycles or ticks, which is how NetLogo identifies the conclusion of a cycle, in this case economic. This code's only action is to generate a random state in each one of the agents, which may be positive or negative; if it is positive, it means the firm has specific assets (knowledge, capital); while if it is negative, it is not a candidate to a possible relation of cooperation with another firm. This way, over the first iteration and once the first states are randomly generated, the first links are established. This is the easiest way to simulate a process of cooperation; though it lacks feedback elements, thresholds and other variables that characterize a process of this nature.

The graph displays the results of running the model for eight iterations. This way, over the first cycle, seven out of the eight firms are yellow, i.e., they take an innovative mode of action, and only one is red, individual innovation mode of action. Over this cycle, only one cooperation relation is developed between two firms. In the second cycle, we notice one new cooperation relation, and three firms have adopted routine mode (indicated in blue). The third cycle indicates four cooperation relations. In the fourth, half of the firms choose routine mode. Over the eighth cycle, we have 14 links, whereas the rate of the selection of the mode of action remains constant.

The results observed in figure 4 reflect a simplification regarding the large amount of variables that concur in a cooperation project, from considering the transaction costs as it is actually developed to the changes that have to be made in algorithmic terms to alter variables such as trust (tr) and amount of knowledge (kr), which should change with each successful cooperation, or the way in which such cooperation affects demand and / or market share.

Conceptually, ^{Beckenbach's et al. (2009)} model is nothing but an iterative process that simulates cooperation between the various agents in a sector of a RIS, which implies, among other things, knowledge transference and absorption, as well as generation of products and/or innovations result of such cooperation relations. Over the course of this iterative process, the knowledge inventory of each firm can improve depending on the model's various feedback factors; as iterations advance, trust can increase or decrease depending on whether collaboration relations are successful.

The parameters used for each one of the agents' variables are based on the empirical data gathered through the research questionnaire. Owing to the sample selection, which focused on firms innovating in biotechnology, the agents' typology is similar, and corresponds to a firm with high level of knowledge specificity and moderate trust capacity, nevertheless willing to cooperate and transfer knowledge, linked to firms oriented to services, not to the industrial.

The amount of knowledge transferred over each iteration depends on both the level of trust and cooperation propensity of each firm, and it is thus that these factors affect the general capacity of RIS to produce cooperation relations and innovations in the market. The generation of a full code for such conceptualization is only the beginning and leaves the door open to a vast field of research, which allows testing the influence of various variables on a RIS efficiency.

Although these considerations surpass the scope of the present text and it is expected to carefully address them in the future, we consider that the code generated is sufficient to illustrate the powerful tool social sciences have in the incorporation of CAS-based research, which allows modeling all sorts of systems aided by a programming language, relatively simple and widely used by the scientific community.

CONCLUSIONS AND RECOMMENDATIONS

In Latin America there is still much to do as regards homogenization of data on innovation measures. In Sonora, very few are the firms engaged in biotechnology, and even fewer those willing to provide information, so the analysis of the data available is a merely exploratory exercise.

Most of the firms under study identify themselves as innovative, prone to cooperative innovation, owners of highly-specialized human capital, possessors of a high level of knowledge; which, by and large, sets relatively ambitious goals, in terms of revenues, market penetration and expansion.

The forgoing clashes with some observations made along this research such as the marked reticence to cooperate by answering a questionnaire among a large group of agents, including academics and entrepreneurs, and a sort of mistrust from governmental actors. They had to be excluded from the set of data in this model, so a future research that incorporates them is recommended.

Capabilities are latent; firms such as Rubio Pharma, Agropro and Galaz Science and Engineering, engaged, respectively, in pharmacology, precision agriculture and production of biomedical instruments, have strong connections abroad, and significant links with research centers and centralized institutions, local HEI and foreign universities .

The scientific community must be attentive to the evolution of the behavior of the RIS, studying this and other emerging sectors, enlarging the available datasets and providing relevant economic information so that to the extent possible it becomes an agent of change in its growth and consolidation

In the simulation model set up by ^{Beckenbach et al. (2009)} in Germany, cooperative innovation has an important role in knowledge dissemination at regional level. Not only do the agents' differenced behaviors provide them with heterogeneous chains of knowledge, but the knowledge on the region as a whole is different every cycle and the importance of each field knowledge varies over these.

In general, for ^{Beckenbach et al. (2009)}, the number of cooperative innovations and dissemination of knowledge among the agents are the most important indicators of the RIS performance. The agents that choose the mode of cooperative innovation are the core of it and this mode of action is the source of multiplex relations both in the market and hierarchy. This analysis shows that the parameters for cooperative innovation reveal a "network landscape" behind the dynamics observable in a RIS.

The broadening of the perspective toward such landscape sheds light on the conditions for a good performance; thereby, such simulations may be a starting point for the bottom-up improvement of RIS –including networks– unlike the usual top-down optimization perspective in network research and that of the “master minds” behind a RIS.

Future research of this nature in Sonora and other regions of the world on RIS in formation will require the incorporation of new variables for the prospective study of various productive sectors and the firms comprised in them, especially those related with scientific and technologic cooperation and collaboration, in addition to ensure technical clarity and replicability.

In like manner, the pressing need of collaboration between social sciences and computer sciences is evinced, as they are necessary as a bridge to generate multiagent simulations based on social and economic theories taken to the algorithmic sphere.