Reviewed by
Robert Holmes
Qforma, Santa Fe
Can complexity research improve public policy? “Yes” say the complexity scholars who have contributed to this book from Brazil’s Institute for Applied Economic Research. But are the authors justified in this positive response?In his preface, Scott Page neatly juxtaposes the apparent benefit of the complexity view with the fact that few involved in setting and implementing policy seem to take it: “Given the complexity of the political and bureaucratic processes that generate policies and the complexity of the systems within which most policies are applied, it would seem that complexity’s relevance should go without saying. Yet, that’s not the case. The patchwork of models, concepts, and ideas that comprise the field of complexity studies rarely enter into policy discussions and when they do, they primarily engage at the fringes.”
So why is complexity not being used? And can this book improve the situation?
When a policy maker—or indeed any stakeholder who has to make decisions and take actions on the results of a model—is presented with a new modelling paradigm, he or she has a simple question: “Does the model work?” Typical follow up questions are “Does it work better than what I have already? How can I tell?” If the answers to these questions are “It should. Maybe. You can’t”, then that policy maker is likely to be somewhat cautious about adopting the new methodology.
This, then, is what I want from this book. I want to know how to build a valid, verifiable model. I want guidelines on how to choose between competing models. I want case studies that illustrate how these valid models have yielded insights into the operation of specific policies, insights that would otherwise not have been uncovered. I want examples where the impact of these novel policies was monitored.
Unfortunately, the book fails to provide this.
The book is divided into three parts: theory and methods, policy objects as complex systems, and case studies (primarily from Brazil).
The theory section is comprehensive and name-checks the usual panoply of techniques (game theory, machine learning, cellular automata, ABMs, etc.) and the usual range of tools (NetLogo, Swarm, NetworkX et al.). But it reads like the theory section of any complexity textbook that has been published in the last 15 years. That is disappointing. It gives the impression of a discipline that is stagnating: one that is struggling with the same issues as it always has and not making headway, not opening up new avenues of thought or exploration.
It is not all bad news though. Matters improve in Gentile et al. (chapter 4) and Sichman (chapter 5) where their complementary descriptions of how a model should be developed are realistic and provide a sensible template. Reassuringly both make proper mention of the importance of verification and validation but again, I was disappointed in the absence of specific guidance on how to do this crucial step.
Part II is something of a mixed bag. It takes four specific objects of public policy— cities, economies, the environment, and societies—and argues that they are examples of complex systems. This is not a hard argument to make. The same points get made four times over: the system shows heterogeneity, self-organization, path dependence, and adaption; therefore it is a complex system. True, but I am not sure that this argument merits 150 pages of text.
The real value of these particular papers is that they act as gateways to the broader research on the specific policy object (Tessone, in chapter 7, is particularly strong on social systems). Proper consideration is also given to the difficulty of calibrating models (see Dawid’s chapter 9) along with the troubling conclusion “… systematic estimations of the models put forward in the literature are at this point largely missing…”. When advocates of complexity make this statement, is it any wonder that policy makers are reluctant to use their techniques?
Part III suffers from too much focus on case studies from Brazil as frankly they are not case studies. They are proof-of-concept pre-case studies, the conclusion of which is typically “this model should work” rather than an actual assessment of whether it does. Indeed the authors acknowledge this deficiency. For example, Mueller (in chapter 12) states “Not only are there no cases of public policies explicitly using these approaches but even in the area of research and academia there is still very little being done.”
Bar-Yam’s paper (chapter 11) is the highlight of part III and he provides an impressive breadth of applications. However this breadth forces the consideration of each individual application to be shallow. Answers to the key questions—how do I calibrate a model, how do I validate it, how do I select between competing models—are as elusive as ever.
Return to Contents
of this issue