Practical Applications for Complexity Theory in Marketing Research

Practical Applications for Complexity Theory in Marketing Research

Practical Applications of Complexity Theory in Marketing Research

Rosanna Garcia

Northeastern University

College of Business Administration

202 Hayden Hall

Boston, MA 02115

617-373-7258

1.0 Introduction Few marketing researchers have used the theories/philosophies behind complexity theory to explore the market place. The systemic approach, which is a foundation of complexity theory, is rarely taken by marketing researchers. Traditional marketing research studies take a micro or macroeconomic viewpoint of the market place. Consumers are studied as separate entities that make decisions on which products to purchase, and firms are studied as another type of entity that make decisions on what products to introduce to the marketplace. Consumer marketing research has more of a focus on the psychological perspective of consumer decision making whereas managerial researchers take more of an organizational perspective. Rarely are the two approaches studied together. Although this is a simplistic mindset to marketing research, it does exemplify the mutually exclusive approach common to marketing researchers. Complexity theory, with its systemic perspective certainly has a place in the social sciences, so why not in business applications such as marketing studies? Are there any practical applications for complexity theory in researching marketing issue?

Another valid question would be, “Why hasn’t marketing researchers embraced complexity theory?” There are several good reasons. Foremost, as previously noted, marketing researchers specialize in either a micro or macroeconomic perspective of the market place. This is primarily due to the expense/complication involved with collecting data. Gathering data from consumers is cumbersome and costly so adding another layer –firms– only adds to the enormity of the task. Secondly, business researchers are rarely trained in the ‘hard’ sciences and are not familiar with simulation modeling and software programming tools necessary to adequately explore complex systems.

But there are numerous reasons why complexity theory is useful in marketing research. For the researcher willing to overcome the learning curve of computer programming, it provides a method for easily evaluating how ‘micro’ elements (consumers) emerge into ‘macro’ elements (marketplaces). It provides an inexpensive vehicle to simulate ‘real’ heterogeneous consumers in ‘real’ marketplace scenarios. Data that had previously been difficult to collect, such as social networks, can now be studied at a micro-level and scaled to macro-level simulations. It also provides a practical and economical method to confirming intuition, such as existing theories regarding word of mouth and tipping points.

In this paper we focus on the methodology, agent-based modeling (ABM), as a vehicle for studying complex adaptive systems. We first introduce the benefits and usefulness of ABMs. We put a particular emphasis on new product development and innovation issues as they particularly lend themselves to agent-based modeling. We also provide an example of a simple algorithm to demonstrate ABMs in a competitive R&D environment. We conclude this paper with issues and future research focused on NPD and innovation using ABMs.

Agent-based modeling approaches are commonly used to represent individual actors (or groups) in a social dynamic system. 'Agents' can be consumers which act as autonomous decision-making entities that interact with each other and/or with their environment based on a set of rules. Repetitive interaction between agents is a basic feature of agent-based modeling so it is ideal for simulating dynamic market places. From these micro-scale interactions, aggregate macro-scale behaviors may emerge. From a new products’ perspective, ABMs has been used to model the diffusion of innovations (Young, 1999, Goldenberg, et al., 2002), innovation networks (Gilbert, et al., 2001), and technological forecasting (Bhargava, et al, 1993).

For example, take a simplistic (although highly unlikely, but illustrative) rule for diffusion. Imagine a marketspace where there is only one rule that all agents ‘follow’. The rule is that each agent looks at its four neighbors (the ones immediately to its left, right, above and below), and if an odd number of them (that is, 1 or 3) have adopted the innovation, it adopts the innovation. If an even number of neighbors (2 or 4) has adopted the innovation, it does not adopt the innovation. Imagine how you would expect this market to look. We obtain the following pattern as shown in Figure 1. This simplified ABM, where only two states exist (on and off) and where agents only interact with immediate neighbors, is based on cellular automata. By modeling real-world systems ABMs can potentially provide valuable information about the dynamics of the system that may not be intuitively obvious.

Figure 1: Simple Model of Diffusion using – Gilbert’s (2002) Parity Model based

on Cellular Automata

2.0 Benefits of Agent-based Modeling. The benefits of ABM over other modeling techniques are numerous. Shortcomings, of course, also exist and will be discussed in the final section. However, it is our belief that that the benefits of this methodology outweigh the shortcomings if the ABM mindset is understood. The role of simulation is not to create an exact facsimile of any particular system or environment but to assist in the exploration of the consequences of various contingencies. Or in other words, simulations should be used as a tool for the refinement of theory (Bonabeau 2002). By simulating real world behavior that may be difficult to capture in static models, the ABM approach focuses on how processes evolve over time and how policies might be changed to affect the outcomes of an evolving system. This methodology accords with Axelrod's (1997) description of the value of simulation. If ABMs are taken to be a learning tool to guide intuition in refining innovation theories, they can be useful to the new product development researcher.

ABMs are best suited to domains where the natural unit of analysis is the individual (consumer, firm, employee, etc.) and when both micro-level behavior of individuals and macro-level patterns emerge from the interactions of these individuals. Modeling ABMs require understanding the behaviors of the agents and translating these behaviors into rules for agents to act upon in the modeled environment. One can easily study individual agents (micro-level), subgroups of agents and aggregated agents (macro-level) behaviors with different levels of rules coexisting in a single model. Due to this disaggregation, dynamic environments can be created where agents enter and exit the system (i.e., under performing firms leave the marketspace or new competitors enter the marketspace). The phenomena of interest are the impact at the macro-level from changes occurring at the micro-level. Tesfatsion (2002) provides an excellent summary of emergence in the related field of economics, also called agent-based computational economics. She refers to emergence in an evolving system of interacting agents as ‘growing economies from the bottom up’.

ABMs are useful when individual behavior is nonlinear such as when learning and/or adaptation occurs on a microscopic level. For example, models where individual behavior exhibits memory, path-dependence, hysteresis, non-markovian behavior, or temporal correlations are easily modeled with ABMs. Stochasticity can also be applied to the agents' behavior. With ABM, sources of randomness can be applied at the appropriate decision-making level as opposed to entering noise terms more or less arbitrarily, which is typical in modeling stochastic processes.

ABMs also make it easier to distinguish physical space from temporal space, which is useful in modeling social networks of diffusion and global location of innovation project team members (both issues to be further discussed below.) The study of the spatial make-up of social networks has been a major focus of study in recent years (Barabási 2002, Watts and Stogatz 1998, Valente 1995). Watts argues that local interactions between network members have global consequences, but that the relationships between local and global dynamics depend on the network's structure. Garcia, et al (2003) demonstrated how different types of social networks can affect the rate of diffusion. They tested four different network structures for effects on the diffusion process of technologically-advancing products: a random network (Erdős and Rényi 1959), a cellular automata network (Goldenberg, et. al 2002), a small world network (Watts and Stogatz 1998) and a scale free network (Barabási and Albert 1999). However, it should be noted that many ABMs are not spatially dependent.

One of the reasons underlying ABMs’ popularity in other social sciences is its ease of implementation. ABMs are easier to construct than other analytic models, albeit, some computer programming knowledge is useful. ABMs do not require extensive knowledge of ordinary differential equations or other analytical modeling techniques. It is reasonably easy to create an ABM using Excel or Matlab. Additionally, several open system software programs are available for download from several sites (Netlogo, Repast, etc.). More sophisticated ABMs sometimes incorporates neural networks, evolutionary algorithms, genetic algorithms, and other learning techniques to allow realistic learning and adaptation and are, therefore, computational more complex (Bonabeau 2002). These complex models, however, are not the norm.

To summarize, ABMs are useful in marketing research:

·  when both macro and mirco-levels of analyses are of interest (e.g., adoption and diffusion of innovations).

·  when social systems cannot easily be described by differential equations but by what-if scenarios (e.g., organizational cultures).

·  when emergent phenomena may be observed (e.g., emergence of innovations).

·  when co-evolving systems interact in the same environment (e.g., competitive markets).

·  when learning or adaptation occurs within the system (e.g., R&D collaboration).

·  when physical space and temporal space is of interest (e.g., supply chain networks).

·  when the population is heterogeneous or the topology of the interactions is heterogeneous and complex (e.g., social networks).

3.0 Possible Applications. The possible applications for ABMs in the study of new product development and innovations have not yet been fully discovered. We provide a few examples of possible research issues that could be addressed by agent-based models. These include, but are by no means limited to:

1.  Diffusion of innovations

·  effects of network externalities,

·  word-of-mouth networks,

·  modeling tipping points,

·  social networks and viral marketing.

2.  Organizations

·  innovation networks & collaboration,

·  co-evolution of competitive strategies,

·  R&D 'emergence' of innovations,

·  portfolio management,

·  innovation strategies & external environmental influences.

3.  Knowledge/Information flows

·  supply chain networks,

·  innovation/R&D collaboration (inter and intra-organizational),

·  technology transfer (inter and intra-organizational & to/from customers-lead users),

·  strategy planning (organizational).

Diffusion of Innovations. Modeling the diffusion of innovations in social networks has been the most common application for ABMs in marketing studies to date (e.g., Bonabeau 2002, Guardiola et. al 2002, Young 1999, Nyblom, et. al. 200x, Goldenberg, et al. 2002). It is fairly straightforward to translate the prototypical diffusion model originating from Bass (1969) into dynamic simulations of heterogeneous consumers using ABMs. Consumers are modeled as agents who make adoption/purchase decisions based on 'word of mouth' influences from local interactions with other agents. 'Other' agents can be modeled as influential consumers, seeding agents (hired guns to positively influence potential adopters), firms with marketing campaigns or other types of mass media. This is analogous to the Bass model where there are innovators and imitators, each adopting an innovation at different times. Unlike the seminal Bass model, ABMs allow the introduction of heterogeneity with respect to initial perceptions, adoption thresholds, and even individual responsiveness to information so that more realistic marketspace environments can be simulated.

ABMs can readily incorporate the idea of influencing agents. Dawkins (1976) introduced the idea of 'memes' as contagious information patterns that infect a human's mind to alter their behavior causing them to propagate the patterns among a population. Typical patterns of memes are slogans, catchy-phrases, melodies, icons, inventions and fashion trends. Seeds, or carefully planted influencers, can initiate the propagation of the "memes" (rhymes with genes). The connection between memes and effective advertising campaigns is evident. Remember the "Where's the Beef?" campaign? Gladwell (2000) examined the role of memes in diffusion as the search for 'tipping points' and Rosen (2000) examined them as 'buzz' or word-of-mouth influence.

Recent ABM studies have also examined the effects of different types of social networks on diffusion patterns (Valente 1995, Janssen and Jager 2002). Different types of social networks include small world networks (Watts and Strogatz 1998), random networks (Erődos and Rényi 1959), and scale-free networks (Barabási and Albert 1999). These network models have roots in social systems where most people are friends with immediate neighbors/family/co-workers, yet are also connected with people located some distance away (spatially or temporally). Scale-free networks have been used to explain the effectiveness of viral marketing (Barabási 2002).

Organizations. Modeling organizational structures and innovation strategies has been other prominent applications for ABMs (Dawid 2001, Debenham and Wilkinson 2003, Garcia and Calantone 2003, Gilbert, et. al. 2001). Agents in these models represent firms, policy actors, research labs, etc. each generating knowledge bases, which lead to innovations. The dissemination of knowledge/information between agents can easily be modeled to represent collaborative or competitive strategies. Gilbert et al. considered innovation networks as evolving from the dynamic and contingent linkage of heterogeneous units each possessing different bundles of knowledge and skill. In order to study these co-evolutionary types of systems, the Self-Organizing Innovation Network (SEIN) project was established by the European Commission Framework 4 Programme. Several studies focusing on international innovation networks have come from this project (SEIN 1999).

The increasing orientation towards internationalization with regards to innovation processes lead to issues not only of structure but also of time and space. Internationalization of product development teams result in globally located team members working on the same product from different localities. This global location requires coordination across time zones and geographics, as well as across cultures. ABMs can be used to answer how cooperation and communication between groups might be established to optimize design times and time to launch. Since ABMs can be modeled as heterogeneous individuals even cultural issues can be modeled into these studies. Collaboration (Gilbert, et. al 2001) and the strength of weak ties in linking innovating partners are easily modeled using agents each possessing their own decision-making criteria.