Order this book
César García-Díaz and Diemo Urbig
University of Antwerp, Belgium
The book splits into two parts, one on theory and one on empirical studies and validation of agent-based models. In the first part, the authors summarized some old and new issues related to the knowledge economy and the role of innovations, which are related to defining knowledge and describing basic mechanisms of its diffusion (chapter 2). They continued briefly presenting some selected models, including epidemic models, game theoretic models, and in a more detailed fashion they discussed knowledge flow models (chapter 3). Following this overview, the authors introduced their own firm-level agent-based model of knowledge sharing and resulting diffusion and provided a first analysis (chapter 4). In the second part, the authors started with an overview about empirical studies on knowledge flow and explicitly addressed how knowledge flow was and could be measured (chapter 5). They discussed the agent-based modelling approach and quickly turned to discussing the validation of agent-based models in general (chapter 6). The authors proceeded by presenting two of their own efforts in validating agent-based models and finally used a case study to validate the model presented in chapter 4 (chapter 7). The final chapter briefly summarized the book and identified some opportunities for future research (chapter 8).
That said, it is worth noting that the book's subtitle "Modelling complex entrepreneurial behaviours" is misleading, because the authors did not make use of the extensive literature on entrepreneurship and corresponding efforts on modelling entrepreneurial behaviour, e.g., journals such as the Journal of Business Venturing and the Entrepreneurship, Theory & Practice, which both are dedicated to entrepreneurship research (see for instance Levesque 2004 on mathematical and formal modelling of entrepreneurial behaviour and McKelvey 2004 on agent-based simulation and complexity research). By missing this research and the related empirical studies, the book ignored potential sources that could have been used for validating models, which is one of the major issues in this book.
At the core of the book (Chapter 4) there is a very interesting agent-based model of knowledge sharing that shows how firms acquire and share skills (knowledge) until they successfully reach an innovation. The authors assumed a directed network of skills, with all predecessors of a skill needed to be acquired in order to be able to achieve a subsequent skill. This model nicely reflects the path-dependency of skill acquisition. Further, innovations were predefined combinations of skills. The authors differentiated between incremental and fundamental innovations by the way the corresponding sets of skills were generated. For incremental innovations, one skill in a skill set representing an innovation was replaced by a skill that was a successor in the skill network. Fundamental or radical innovations represented skill sets that were independent of other innovations. The authors assumed that firms innovated by selecting a not yet implemented innovation from the pre-specified list of potential innovations; this was implemented if the firm possessed all necessary skills. After an initial skill development, firms were capable of acquiring additional skills only through partnering with other firms. They partnered if they could not innovate on their own. If firms cannot innovate individually, then they searched their acquaintances so as to pool the necessary skills. If they successfully innovated in a partnership, the shared skills were diffused to the partner. Therefore, the network structure of potential partners and its relation with firms' innovativeness were at the core of the authors' research question.
It is worth saying that the model of skills and innovations presented in this book is very interesting and we believe that it has potential for further research. Unfortunately, at the present stage, the model remains under-explored. To begin, the authors did not give much explanation or justification about the parameter values used in the simulations. For instance, they assumed that the number of radical and incremental innovations was the same. The authors acknowledged that their full specification put them into the need of "adding assumptions as much as it involves abstracting from a body of theoretical knowledge" (page 107). In experimental research, these assumptions are referred to as auxiliary assumptions (Croson and Gächter 2010). An important step in simulation (and experimental) analysis is to understand to what extent these auxiliary assumptions influence the results. Sensitivity analysis is a mean to gain such insights (e.g., see Richiardi et al. 2006). Comprehensive sensitivity analyses, however, are missing in the book. Further, the book authors adopted a sample size equal to 100 runs and reported the mean results without assuring statistically meaningful outcomes (i.e., mean values reported with confidence intervals, as well as correlation values with statistical significance).
Another criticism is that the book offers a not very comprehensive review of the literature about theoretical models of innovation diffusion. Since this book was intended as a piece of work dedicated to a complexity-based approach, it is curious to see that the authors did not mention nor linked their effort to the NK model literature, e.g., the explanation of innovation dynamics provided by NK models (e.g. Levinthal 1997; Frenken 2000; Sorenson, Rivkin and Fleeming 2006). For the intermediate or advanced reader, a more detailed review on innovation diffusion models can be found in Young (2009). Moreover, the book is not embedded into recent research on the role of social network structure in diffusion processes, such as Chang and Harrington (2007), or as Lazer and Friedman's (2007), and Lee, Lee, and Lee's (2006) simulation studies that built on Levinthal's model of firms searching a technology space (Levinthal 1997). In the second part of the book, such limitations become even more obvious when omitting published empirical research directly related to the main topic of book (e.g. Sorenson, Rivkin and Fleeming 2006).
In several parts of the book, the authors show a clear discontent with neoclassical economics assumptions, like sophisticated rationality, the use of representative agents and perfect information. We would like to take the opportunity of this review to argue that the claim to improve neoclassic economics is neither innovative nor a contribution in itself. Strong arguments with simulation modelling exercises that undermine neoclassical economic approaches have appeared in the literature (e.g. Sterman et al 2007) but this book falls short in proceeding in the same fashion. In fact, the book authors fail to acknowledge that some of the criticized claims are outdated, because well-established economic research already acknowledges heterogeneity among actors as well as imperfect information (e.g. see: Parker 2009 on economics of entrepreneurship, Chamley 2005 on economic models of social learning, or Young 2009, on tractable ways to include heterogeneity in mathematical models). Innovative texts need to set themselves apart from recent research and not from outdated models that are merely used as benchmarks, such as old neoclassical economics.
Another criticism is that if agent-based models have the capacity of representing rich non-linear dynamics, this aspect, in our view, has been not fully exploited in the model analysis reported in the book. The authors' major finding is that the density of acquaintances networks and size of the largest component affected the final innovative performance in an industry. The authors explicitly assumed a dependency between network structure, i.e. number of links (equivalent to density), and innovativeness. New links were added exogenously to the firm networks as new innovations were generated. Given the fact that the innovativeness over time dependeded on the initial skill set (especially for early individual innovations) as well as on the set of acquaintances (for joint innovations), and that the growth in the network was affected by previous innovativeness, one can easily hypothesize that the initial network structure was positively associated with the innovativeness in an industry. Instead of general motivations for agent-based simulation, we would have appreciated a more elaborated discussion about the need for simulations to study the relation under consideration and what were the drawbacks and implications of the initial (auxiliary) assumptions. For instance, a question to be tackled with would have been as follows: what are the limits and complexities, e.g. opposing effects, which hinder simple logical reasoning and which finally motivate the use of simulations for this specific research question? Furthermore, market implications are not taken into account in the performance measures introduced by the authors. In respect to previous models suggested in the literature, it seems here that the only thing that really matters is the number of realized innovations, without room for discussing their type, e.g. radical or incremental, or their economic implications.
The book authors provide a very interesting overview of examples where validation of agent-based models is the key issue. The authors succeed in explaining the need for and in describing potential ways of empirically validating agent-based models. This part is, from our point of view, a strength of this book. The authors really get the reader interested in the validation of agent-based models and of their own model in particular. While the authors describe what they did with respect to their own study, some more detailed descriptions of the surveys and procedures as well as their relation to the agent-based model would have helped the reader to develop own empirical studies for validation purposes. Several of the issues relevant for the empirical research were not sufficiently discussed. We wonder if the use of a low technology-intensive industry, such as the organic food production in Foggia, Italy, is representative for model validation (when the core model was supposed to study innovation-driven industries). Additionally, what starts as a validation resulted as a model calibration of initial conditions according to the Foggia organic food sector data, without offering a way to really compare if the empirical innovation mechanisms and patterns were matched by the model and its outcomes. On page 132, and with no convincing argument, the authors said they excluded extra-cluster relations, being these relations very relevant to understand innovation patterns. Further, the authors selected a subset of firms and analyzed only these firms' network. How representative was this set of firms and their network? Firms are likely to have connection with firms not selected. Here, a more in-depth discussion of methodological issues would have been extremely appropriate. Similar to the scepticism about the representativeness of simulation results, reporting experimental results based on 14 players and selecting 32 out of 120 firms in an empirical study raises doubts. What do these 32 firms tell us about the network among the 120 firms? Given the complexity of the system – as claimed by the authors – how can these samples provide a sufficient amount of information in order to validate the simulation runs? While the book provided substantial introductory information, the application to the specific case was less informative. The documentation of empirical efforts as well as the rather small samples and efforts in order to generalize the findings is more suggestive of the empirical study as an example than as a core contribution.
In summary, although we view this book as a very interesting introductory example of agent-based modelling, analysis and validation that raises very interesting questions and provides some interesting insights, we must say that unfortunately it has missed several relevant points.
CHANG, M-H, Harrington, JE, Jr (2007) Innovators, Imitators, and the Evolving Architecture of Problem-Solving Networks. Organization Science 18(4): 648-666
CROSON, R, Gächter, S (2010) The Science of Experimental Economics. Journal of Economic Behavior & Organization, 73(1): 122-131
FRENKEN, K (2000) A Complexity Approach to Innovation Networks. Research Policy 29(2): 257-272
LAZER, D, Friedman, A (2007) The Network Structure of Exploration and Exploitation. Administrative Science Quarterly, 52, 667-694
LEE, E, Lee, J, Lee, J (2006) Reconsideration of the Winner-Takes-All Hypothesis: Complex Networks and Local Bias. Management Science, 52(12): 1838-1848
LEVESQUE, M (2004) Mathematics, Theory, and Entrepreneurship. Journal of Business Venturing, 19(5): 743-765
LEVINTHAL, DA (1997) Adaptation on Rugged Landscapes. Management Science, 43(7): 934-950
McKELVEY, B (2004) Toward a Complexity Science of Entrepreneurship. Journal of Business Venturing, 19(3): 313-341
PARKER, SC (2009) The Economics of Entrepreneurship. Cambridge, UK: Cambridge University Press
RICHIARDI, M, Leombruni, R, Saam, NJ, and Sonnessa, M (2006) A Common Protocol for Agent-Based Social Simulation. Journal of Artificial Societies and Social Simulation 9(1)15: http://jasss.soc.surrey.ac.uk/9/1/15.html
SORENSON, O, Rivkin, JW, Fleming, L (2006) Complexity, Networks and Knowledge Flow. Research Policy 35(7): 994-1017
STERMAN, JD, Henderson, R, Beinhocker, ED, Newman, LI (2007) Getting Big Too Fast: Strategic Dynamics with Increasing Returns and Bounded Rationality. Management Science 53(4): 683-696
YOUNG, HP (2009) Innovation Diffusion in Heterogeneous Populations: Contagion, Social Influence, and Social Learning. American Economic Review, 99(5): 1899-1924
Return to Contents of this issue
© Copyright Journal of Artificial Societies and Social Simulation, 2010